Improperly designed hardware really can catch fire. Maybe all the computers in Hollywood have pentium 4 CPUs and Tesla GPUs with a few Sony Li ion batteries thrown in to give it that extra spark.

Once you get an iPod/iPhone/iMac/mac mini "open" it is actually very easy to service the parts. The smaller devices use little ZIF ribbons that make it fairly easy, the computers use interfaces that are also easy to figure out even if you've never serviced your own computer. Had apple not made it hard to open these devices there would be a lot more people buying replacement parts to fix these devices rather than shelling out for new ones. Apple could have easily kept the same form factors and made these devices user serviceable, it just means there are a couple screws on those shiny back plates.

If you're going to do anything in the way of signal processing you will need much more advanced math skills. You're going to need to know calculus, fourier transforms, laplace transforms along with their discrete versions (dft, fft, z). These are extremely useful, powerful mathematical techniques for analyzing data. The mathematics behind them are amazing and elegant. Examples of their use include the enabling of the 'digital' content age through video and audio compression. That's just one example, they can be used to analyze any data stream. It is amazing using DSP to find data in a sea of noise.

TDP is thermal design power. It is the number that CPU vendors give to system builders to let them know the max power consumption of the processor so that thermal solutions can be properly designed. If the CPU goes beyond the TDP the system will power off unless there is lots of headroom in the thermal solution.

Having worked on these processors at the circuit level(*) I can tell you that your '100W over TDP' number is rubbish.

If you'd like to know more about what happens when chip vendors fudge on this "invalid metric" search for "nvidia bumpgate". If our chips were running at 100W over spec'd TDP we'd have a lot of very unhappy customers.

* yes, I'm an engineer at AMD and I designed major components on the parts discussed ITFA. I did my time at Intel as well.

There's one truly Killer App that's not on there - Mac OS X. The low-level APIs use OpenGL for hardware acceleration on supported GPUs (pretty much any discrete GPU in a late model Mac).

http://developer.apple.com/macosx/architecture/index.html
http://developer.apple.com/graphicsimaging/opengl/
http://en.wikipedia.org/wiki/Core_Image

AMD's chips are on a silicon-on-insulator (SOI) technology, which comes with less parasitic source/drain cap and no junction leakage. Switching less cap means you switch faster. No junction leakage means your part uses less power when it's idle. High-k will be coming soon enough for AMD, the 45nm tech is competing pretty well w/o it at the moment.

AMD uses an exclusive cache architecture, Intel's is inclusive. Exclusive means that a given line exists in only one cache at a given time. Inclusive means that if a line is in one cache, it's in all of them. For an AMD chip add up all the caches on the die to get total cache size. For 45nm quad-core products that's 8.25MB. For an Intel chip take the size of the last level cache, that's your effective total cache. For Nehalem that's 8MB.

You are correct about issuing and higher IPC coming with Bulldozer.

Sorry, but that's incorrect. Photons do not interact with fields inside atoms, they interact with electrons and protons. That is how lenses, photosynthesis, photodiodes, etc works. Photons are messenger particles of the electromagnetic force. Protons and electrons are the "receivers". Photons do not interact with each other except, as integer spin particles (bosons) they can occupy the same quantum state. This allows for the superposition of two photons to increase, decrease, or cancel out the amplitude of an electromagnetic field at a point in space.

Actually, I just gave an explanation of how phased array radar works. Read up on interference.

Electromagnetic fields are made of photons. An incoming photon will be superimposed (interfere) with the photons in an electromagnetic field as it passes through but it's path will remain unchanged. The only way to influence the path of a photon is to bend spacetime, which is the definition of what we call gravity.

It's actually more complicated than that. The six-core chips have the ability to configure up to a quarter of the 6MB L3 cache as a probe filter. This keeps most snoop traffic from reaching down into the L2 and L1 caches of the other cores on the same die and all cores on other die in a multi-socket system. The result is better memory latency and improved memory bandwidth. Here's a link: http://techreport.com/articles.x/16448

For a while it will be proportional to (V^2)*f. However, transistor leakage is exponential with respect to Vds due to DIBL (drain-induced barrier lowering) and the resulting decrease in threshold voltage. Raising the voltage too much is a double-whammy WRT power consumption.

Actually it does have to do with hard disks - you've got to swap to _something_ once those interpreters use all 4GB of physical RAM!

Not all of us like KDE and Gnome. Although they have innovated in some areas, they're otherwise both just attempts to make an MS clone. While this might be the right thing to make "Linux on the desktop" succeed, it's not what some of envision as the future of computing or even how we'd like the current state of computing.

I've been a Linux user since 1996 and I've watched the entire "Linux on the desktop" debacle unfold. It's sad to see that the /. crowd has changed from the attitude of "these MS clones are crap, desktop Linux should be something better" to "ho hum, this is how it should be, why innovate"? Back when this was just "Rob's page" you would've been flamed into oblivion for that public show of affection for KDE/Gnome.

Who are you to tell Google what they can and cannot build? It's about time someone put a face on the Linux desktop other than that of an MS clone. Hopefully it's not just a new window manager but a new window system. X, while great in its day, has run its course. I'd like to something fresh that builds on the concept of using modern graphics hardware to do all the heavy lifting for the GUI instead of clever CPU-intensive hacks on top of Xlib**.

You don't like Google's vision for Linux on netbooks. What's your alternative vision?

** I've written many apps with Xlib. The underlying ideas/primitives that X uses for graphics ops are obsolete so doing anything "cool" (and sometimes useful!) requires using crufty extensions rather than calling routines that are a "natural" part of the system.

While it's relatively new compared to everything on this list, the AMD Opteron, which came out in 2003, will be the face of computing for the foreseeable future. Even now in 2009 AMD's archival Intel is just coming out with integrated memory controllers and high-speed serial direct interconnects. The Opteron also forced Intel to give everyone 64-bit memory addressing in x86 (which Intel wanted to stay in the realm of high-end RISC/Itanium machines).

Opteron wasn't the first chip to have any of these features, but it was the first _x86_ chip to have all of them - making it an affordable "high-end" processor for small businesses and tech junkies. It really was a "world-shaking" product that put AMD on the map. No one expected little AMD to make a splash so big with Opteron (except Intel, which paid off companies to not release Opteron-based products, a55holes!).